Analysis of nanopore signal using a machine-learning technique

Abstract

Techniques for estimating a polymer sequence of a polymer based on a signal produced as a result of translocation of the polymer through a nanopore are described. The techniques may analyze portions of the signal to estimate whether there was a transition in the polymer sequence during each respective portion and which units of the sequence the transition was between. The techniques may comprise operation of one or more neural networks into which data from the signal may be input. The techniques may include generating a plurality of weights for a portion of the signal, wherein each weight is associated with a transition between labeled units of the polymer. The weights may be indicative of a likelihood that a transition occurred between a first of the labeled units to a second of the labeled units within the portion of the signal.

What is claimed is: 
     
       1. A method of high rate sequencing of polymers using a nanopore measurement and analysis system, the method comprising:
 placing a polymer into the nanopore measurement and analysis system; and 
 sequencing the polymer using the nanopore measurement and analysis system at least in part by:
 translocating at least a portion of the polymer through a nanopore of the nanopore measurement and analysis system at a sequencing rate in the range of 10-1000 polymer units per second, wherein the sequencing rate comprises a rate at which the polymer translocates through the nanopore; 
 measuring, using the nanopore measurement and analysis system and at a sampling rate between 100 Hz and 30 KHz, electrical signals generated by the translocating of the polymer through the nanopore, wherein the sampling rate is greater than or equal to the sequencing rate; 
 generating a time-ordered series of measurements based on the measuring of the electrical signals generated by the translocating; 
 organizing the time-ordered series of measurements into a plurality of overlapping subsets of measurements; 
 generating a plurality of feature vectors from the plurality of overlapping subsets of measurements by processing the subsets of measurements using a convolutional neural network; 
 generating a plurality of sets of transition weights from the plurality of feature vectors using a recurrent neural network; 
 generating, using the plurality of sets of transition weights, a Hidden Markov Model (HMM); 
 determining, using the HMM, an estimate of a sequence of polymer units in the polymer; and 
 outputting the estimated sequence of polymer units, 
 wherein the recurrent neural network comprises a bidirectional recurrent layer, the bidirectional recurrent layer comprising:
 a first unidirectional recurrent layer comprising a first plurality of long short-term memory (LSTM) units connected in a first direction; and 
 a second unidirectional recurrent layer comprising a second plurality of LSTM units connected in a second direction, opposite the first direction, wherein outputs of the first unidirectional recurrent layer are inputs into the second unidirectional recurrent layer, 
 
 wherein generating the plurality of sets of transition weights comprises:
 updating state vectors of multiple LSTM units in the first unidirectional recurrent layer based on the feature vectors and state vectors of LSTM units preceding the multiple LSTM units in the first unidirectional recurrent layer; and 
 updating state vectors of multiple LSTM units in the second unidirectional recurrent layer based on outputs of the first unidirectional recurrent layer and state vectors of LSTM units preceding the multiple LSTM units in the second unidirectional recurrent layer; and 
 
 wherein each weight of a particular set of transition weights of the plurality of sets of transition weights is associated with respective first and second labels and is indicative of the likelihood that a transition between a polymer unit having the first label and a polymer unit having the second label occurred within a measurement period represented by a subset of measurements, of the plurality of overlapping subsets, associated with the particular set of transition weights. 
 
 
     
     
       2. The method of  claim 1 , wherein a number of the measurements in a subset of the plurality of subsets of the time-ordered series of measurements is different than a number of a plurality of values of the feature vector. 
     
     
       3. The method of  claim 1 ,
 wherein each of the polymer units in the polymer is one of a finite, known group of polymer units, the group of polymer units consisting of N distinct polymer units, 
 wherein each of the first label and second label is one of a finite, known, group of labels, the group of labels consisting of M distinct labels, and 
 wherein M is greater than N. 
 
     
     
       4. The method of  claim 3 , wherein a set of the plurality of sets of weights consists of M 2  weights. 
     
     
       5. The method of  claim 3 , wherein M is equal to N+1, and wherein the group of labels consists of N labels each corresponding to respective ones of the group of polymer units, and a single label corresponding to a blank label, which represents a lack of a transition within the measurement period represented by an associated subset of the plurality of overlapping subsets of the time-ordered series of measurements. 
     
     
       6. The method of  claim 3 , wherein M is equal to 2×N, and wherein the group of labels consists of N labels each corresponding to a first instance of respective ones of the group of polymer units, and N labels each corresponding to a second instance of the respective ones of the group of polymer units. 
     
     
       7. The method of  claim 1 ,
 wherein generating the HMM comprises determining emission and transition probabilities of the HMM using weights of the plurality of sets of weights; and 
 wherein determining the estimate of the sequence of polymer units comprises determining, using the HMM, a most likely sequence of polymer units within the polymer. 
 
     
     
       8. The method of  claim 7 ,
 wherein each of the first label and second label is one of a finite, known, group of labels, and 
 wherein determining the most likely sequence of polymer units within the polymer comprises determining the most likely sequence of labels of the group of labels using the HMM, and identifying polymer units that correspond to the labels of the group of labels. 
 
     
     
       9. The method of  claim 1 , further comprising measuring a current through the nanopore during translocation of the polymer through the nanopore, thereby generating a current measurement signal. 
     
     
       10. The method of  claim 9 , further comprising digitizing the current measurement signal, thereby producing the time-ordered series of measurements. 
     
     
       11. The method of  claim 1 , wherein:
 each of the polymer units in the polymer is one of a finite, known group of polymer units, 
 each polymer unit of the finite, known group of polymer units has a respective primary label and a respective secondary label, and 
 transitions, within the plurality of sets of transition weights, from any first label to a second label corresponding to the primary label of a particular polymer unit are allowed, and transitions, within the plurality of sets of transition weights, to any second label corresponding the secondary label of the particular polymer are only allowed from a first label corresponding to the primary or secondary label of the particular polymer unit. 
 
     
     
       12. The method of  claim 1 , wherein the sampling rate is between one and ten times the sequencing rate. 
     
     
       13. A system for high rate sequencing of polymers, the system comprising:
 a nanopore measurement and analysis system comprising:
 a nanopore; 
 one or more processors; and 
 at least one non-transitory computer readable medium storing instructions; 
 
 wherein the nanopore measurement and analysis system is configured to sequence a polymer at least in part by:
 translocating at least a portion of the polymer through the nanopore at a sequencing rate in the range of 10-1000 polymer units per second, wherein the sequencing rate comprises a rate at which the polymer translocates through the nanopore; 
 measuring, at a sampling rate between 100 Hz and 30Khz, electrical signals generated by the translocating of the polymer through the nanopore, wherein the sampling rate is greater than or equal to the sequencing rate; 
 generating a time-ordered series of measurements from electrical signals generated by the translocating of the polymer through the nanopore; 
 performing, by executing the instructions stored on the at least one non-transitory medium with the one or more processors:
 organizing the time-ordered series of measurements into a plurality of overlapping subsets of measurements; 
 generating a plurality of feature vectors from the plurality of overlapping subsets of measurements by processing the subsets of measurements using a convolutional neural network; 
 generating a plurality of sets of transition weights from the plurality of feature vectors using a recurrent neural network; 
 generating, using the plurality of sets of transition weights, a Hidden Markov Model (HMM); and 
 determining, using the HMM, using the one or more processors, an estimate of the sequence of polymer units in the polymer; and 
 outputting the estimated sequence of polymer units, 
 wherein the recurrent neural network comprises a bidirectional recurrent layer, the bidirectional recurrent layer comprising:
 a first unidirectional recurrent layer comprising a first plurality of long short-term memory (LSTM) units connected in a first direction; and 
 a second unidirectional recurrent layer comprising a second plurality of LSTM units connected in a second direction, opposite the first direction, wherein outputs of the first unidirectional recurrent layer are inputs into the second unidirectional recurrent layer; 
 
 wherein generating the plurality of sets of transition weights comprises:
 updating state vectors of multiple LSTM units in the first unidirectional recurrent layer based on the feature vectors and state vectors of LSTM units preceding the multiple LSTM units in the first unidirectional recurrent layer; and 
 updating state vectors of multiple LSTM units in the second unidirectional recurrent layer based on outputs of the first unidirectional recurrent layer and state vectors of LSTM units preceding the multiple LSTM units in the second unidirectional recurrent layer; and 
 
 wherein each weight of a particular set of transition weights of the plurality of sets of transition weights is associated with respective first and second labels and is indicative of the likelihood that a transition between a polymer unit having the first label and a polymer unit having the second label occurred within a measurement period represented by a subset of measurements, of the plurality of overlapping subsets, associated with the particular set of transition weights. 
 
 
 
     
     
       14. The system of  claim 13 , wherein a number of the measurements in a subset of the plurality of overlapping subsets of the time-ordered series of measurements is different than a number of a plurality of values of the feature vector. 
     
     
       15. The system of  claim 13 ,
 wherein each of the polymer units in the polymer is one of a finite, known group of polymer units, the group of polymer units consisting of N distinct polymer units, 
 wherein each of the first label and second label is one of a finite, known, group of labels, the group of labels consisting of M distinct labels, and 
 wherein M is greater than N. 
 
     
     
       16. The system of  claim 15 , wherein a set of the plurality of sets of weights consists of M 2  weights. 
     
     
       17. The system of  claim 15 , wherein M is equal to N+1, and wherein the group of labels consists of N labels each corresponding to respective ones of the group of polymer units, and a single label corresponding to a blank label, which represents a lack of a transition within the measurement period represented by an associated subset of the plurality of overlapping subsets of the time-ordered series of measurements. 
     
     
       18. The system of  claim 15 , wherein M is equal to 2×N, and wherein the group of labels consists of N labels each corresponding to a first instance of respective ones of the group of polymer units, and N labels each corresponding to a second instance of the respective ones of the group of polymer units. 
     
     
       19. The system of  claim 13 ,
 wherein generating the HMM comprises determining emission and transition probabilities of the HMM using weights of the plurality of sets of weights; and 
 wherein determining the estimate of the sequence of polymer units comprises determining, using the HMM, a most likely sequence of polymer units within the polymer. 
 
     
     
       20. The system of  claim 19 ,
 wherein each of the first label and second label is one of a finite, known, group of labels, and 
 wherein determining the most likely sequence of polymer units within the polymer comprises determining the most likely sequence of labels of the group of labels using the HMM, and identifying polymer units that correspond to the labels of the group of labels. 
 
     
     
       21. The system of  claim 13 , further comprising a measurement unit configured to measure a current through the nanopore during translocation of the polymer through the nanopore, thereby generating a current measurement signal. 
     
     
       22. The system of  claim 21 , wherein the measurement unit is further configured to digitize the current measurement signal, thereby producing the time-ordered series of measurements, and to provide the time-ordered series of measurements to the analysis unit. 
     
     
       23. The system of  claim 13 , wherein:
 each of the polymer units in the polymer is one of a finite, known group of polymer units, 
 each polymer unit of the finite, known group of polymer units has a respective primary label and a respective secondary label, and 
 transitions, within the plurality of sets of transition weights, from any first label to a second label corresponding to the primary label of a particular polymer unit are allowed, and transitions, within the plurality of sets of transition weights, to any second label corresponding the secondary label of the particular polymer are only allowed from a first label corresponding to the primary or secondary label of the particular polymer unit. 
 
     
     
       24. The system of  claim 13 , wherein the sampling rate is between one and ten times the sequencing rate.